[evla-sw-discuss] Controlling the baseband slope correctors

[Barry Clark]

To set the baseband slope correctly, the least complicated approach
is to use the Wide Band Correlator on the station board to see the
overall shape of the spectrum, and have it report the correction
needed to the Executor, which would then forward it to the slope
corrector in the T304 modules.

This is because the required slope is a function of antenna, band,
and location within the band.  To do it by tabulation would require
maintenance of a very large table which would change from time to
time as receivers are serviced or replaced.  (The slope required
may also be a function of which T304s are installed, although these
are believed to be much more uniform than the receivers.)

The Executor could handle this information in the same way it currently
handles reference pointing, phasing, and, indeed, setting T304 levels.

What the Executor would like to see is just an rms and slope from
the WBC, one for each baseband in the array.  I suggest the following
XML descriptor.

   <xsd:complexType name="WbcDataType">
     <xsd:sequence>
       <xsd:element name="rms" type="xsd:float" />
       <xsd:element name="slope" type="xsd:float" />
     </xsd:sequence>
     <xsd:attribute name="antId" type="xsd:string" />
     <xsd:attribute name="baseBand" type="xsd:string" />
     <xsd:attribute name="avg" type="xsd:double" />
     <xsd:attribute name="time" type="xsd:dateTime" />
   </xsd:complexType>

These would probably come packaged in an envelope, either for a lot
of antennas if they come from some collector in the correlator, or
with just a couple if they come directly from the station board.

Worst case is eight basebands from 28 antennas, 224 elements in all.
The whole works might fit in a 16K datagram.  I think a reasonable
averaging time (avg above) would be 10 seconds.  (The consideration
is only to keep the traffic at a low level while still getting answers
on a script-type timescale.)  I guess CM will need to send this
averaging time to the StB.

These need only be sent when in three bit mode.

The rms above is just the lag zero autocorrelation, divided by
averaging time and suitably scaled.

The slope is a little more difficult.  If computation is not a
consideration, you calculate the spectrum and see which slope
corrector works best, calculating the goodness of fit of
    a*10^{0.1*n*(f/bw)}
for, say, n in the range -10 to 10 and reporting the best n.
If calculation is a problem, a reasonable guess can be made directly
from the lags.  (The lag beyond which the correlation is low is
a measure of the magnitude of the slope; whether there are nulls
before then gives the sign.)